Unit overview

Where the majority of machine learning work actually happens

T-08 of Unit 1 established that data preparation and exploratory analysis typically consume 40–60% of a real project's time. This unit is where that work begins. Unit 1 was about ideas; Unit 2 is about the raw material those ideas operate on — its types, its quirks, its failures, and the ways practitioners shape it into something a model can learn from.

Learning outcomes

By the end of this unit you will be able to

Classify a dataset as structured, unstructured, or semi-structured, and identify the implications of each type for modelling.
Distinguish features from labels in any dataset, and identify which columns would be appropriate inputs and which would be targets.
Diagnose common data quality problems — missing values, outliers, duplicates, inconsistent formats, and mislabelled examples — and explain the consequences of each.
Select an appropriate preprocessing strategy for a given data quality problem, and justify the choice.
Explain the difference between preprocessing (making data usable) and feature engineering (creating new information), and apply both appropriately.
Split a dataset into training, validation, and test partitions, and explain the role of each partition in the modelling process.
Recognise how data decisions made in this unit connect to the failure modes from T-07 — particularly data leakage and spurious correlation.

The shift from Unit 1

Unit 1 was entirely conceptual. You did not touch a dataset. Unit 2 is the first unit where you work through concrete examples with real structure. You will not yet be using Python — that begins in Unit 3 — but you will be reasoning about the kinds of decisions that every data practitioner has to make, using small examples you can work through on paper.

Prerequisite reminder

This unit assumes you have completed Unit 1, particularly T-05 (where features, labels, parameters, and loss are defined) and T-07 (where data leakage and spurious correlation are introduced). Both concepts return repeatedly here.

Tutorials

Nine tutorials, worked through in order. The first four establish the vocabulary (what kind of data exists, what features and labels are). The next four apply that vocabulary to the practical problems every dataset presents. The final tutorial reviews the unit.

T-01

Data in the Real World

Where data comes from, how it is collected, and why real datasets are never as clean as they look.

T-02

Structured Data

Tables, rows, and columns — the format most classical machine learning operates on.

T-03

Unstructured Data

Images, text, audio, and video — the raw material of deep learning, and the harder problem.

T-04

Features and Labels

A return to the four terms from T-05, now grounded in real datasets and real columns.

T-05

Data Quality and Its Consequences

Missing values, outliers, duplicates, and the ways poor data quality turns into poor models.

T-06

Data Preprocessing

Handling missing values, scaling features, encoding categories, and preparing data for a model.

T-07

Feature Engineering

Creating new features from existing ones, and why this is often more valuable than model choice.

T-08

Training, Validation, and Test Sets

Why we split data, how to split it correctly, and what can go wrong when we do not.

T-09

Unit Review and Self-Assessment

A structured review of the unit and a short quiz to check your understanding.

How to work through this unit

Unit 1 was about ideas. Unit 2 is about the raw material. The habits that matter here are different again — closer to inspection than interpretation.

Work through the examples by hand
Unit 2 uses small datasets you can inspect on paper. Do not skip the manual exercises — the point is to build the habits of attention that real data work demands.
Look at datasets you already have
As you work through the unit, examine real datasets you have access to — a spreadsheet from work, a public dataset, even the CSV export of your bank statements. The concepts only become concrete when applied to real, messy data.
Ask what each column actually means
Half of data preparation is understanding what each field represents, when it was recorded, and who recorded it. This is exactly the discipline that catches data leakage before it causes problems.
Finish with the unit quiz
The quiz at the end of T-09 is short and diagnostic. Score it honestly before looking at the feedback.

Assessment in this unit

Unit 2 contributes one unit quiz to your final grade. The assignment that covers this unit's material is released in the next module, so the quiz is the primary assessment here.

Quiz 2 Data Fundamentals — 12 questions, counts toward your quiz grade Go to T-09

Assignment 1 is coming

Assignment 1: Data Exploration and Preparation will be released at the start of Module 2 and is due in week 4. It draws directly on the material in this unit — everything it asks you to do is a practical application of the concepts you will cover here. Treat the tutorial exercises in this unit as preparation for it.

Looking ahead

Unit 3 is the final unit of Module 1, and it is where you get your hands on the tools. Everything you have reasoned about in Units 1 and 2 becomes practical in Unit 3, where you will learn to use Python, Jupyter, Pandas, and Scikit-Learn to explore, clean, and prepare real datasets. The unit ends with your first complete end-to-end workflow — the same kind of workflow that Assignment 1 will ask you to perform.